CN102499675B

CN102499675B - Feedback system random resonance intensification method of electro-corticogram signal

Info

Publication number: CN102499675B
Application number: CN 201110331467
Authority: CN
Inventors: 范影乐; 王海玲; 赵磊; 郭芳芳; 陈金龙
Original assignee: Hangzhou Dianzi University
Current assignee: Yunfeng Pipe Industry Co Ltd
Priority date: 2011-10-27
Filing date: 2011-10-27
Publication date: 2013-09-18
Anticipated expiration: 2031-10-27
Also published as: CN102499675A

Abstract

The invention relates to a feedback system random resonance intensification method of an electro-corticogram signal. Different from a traditional intensification method of the elctro-corticogram signal, the method is advantageous to improvement of the performance of intensifying the electro-corticogram signal through a random resonance mechanism under noise interference with certain strength. In consideration of the characteristics of electrophysiological signals such as the electro-corticogram signal, a FitzHugh-Nagumo model in accordance with the neuron electrophysiological characteristics is adopted, and a feedback element is introduced to a one-way neuron network structure, so that the method is more in accordance with the complicated interconnection relationship among neurons in a neuron system. Compared with the traditional noise filtering method, the method can better restore and intensify the electro-corticogram signal.

Description

A kind of feedback system accidental resonance Enhancement Method of cortex EEG signals

Technical field

The invention belongs to biomedical engineering field, relate to the processing method that a kind of cortex EEG signals strengthens, be specifically related to a kind of faint cortex EEG signals real time enhancing method based on neuron Feedback network model random resonance mechanism.

Background technology

Because the cortex EEG signals easily is subject to the interference of various electricity physiological signals and other noises, the cortex EEG signals that therefore detects is fainter, and Noise is larger, and this studies EEG signals for us and has a great impact.Effectively eliminate the adverse effect of noise, the EEG signals that remains with simultaneously usefulness becomes particularly important.Traditional noise for harmful is mainly taked the method for filtering, the priori of interfering signal, the cortex EEG signals signal to noise ratio that collects is lower, if removal noise, EEG signals can be subject to very large infringement, even EEG signals can be used as together filtering of noise, so that the EEG signals distortion can't be restored.Because random resonance mechanism can be in harmonious proportion relation between nonlinear system, signal and the noise, noise energy is shifted to signal energy, therefore random resonance mechanism thinks that the existence of noise is significant, the cortex EEG signals that suitable signal to noise ratio is lower in some sense.The nonlinear system of present existing accidental resonance derives from the physics abstract models such as bistable system usually, and it realizes the enhancing of EEG signals by EEG signals, noise information and the triangular accidental resonance of nonlinear system.But these physics abstract models are too idealized, and whether its electro-physiological signals to this quasi-representative of brain electricity is fit to, and does not have sufficient foundation.Therefore the present invention proposes to adopt FitzHugh-Nagumo (FHN) neuron models that can truly reflect the neuron electrophysiological characteristics, and forms the network-feedback structure with being connected to each other between the simulation cerebral neuron; To be subject to the input of the low signal-to-noise ratio cortex EEG signals of various interference as aforementioned feedback neural metanetwork model, utilize random resonance mechanism to realize the enhancing of weak signal.

Summary of the invention

The present invention is based on the FHN neuron models, a kind of random resonance mechanism based on the neuron Feedback network model is provided, realize the enhancing of faint cortex EEG signals with this.Feedback element has been added on basis at the FHN neural network model, avoids unidirectional control and the unstability of neural network model with this, and the realization output signal is improved the accidental resonance performance of FHN neural network model for the regulating action of input signal.

The inventive method may further comprise the steps:

Step (1) obtains the cortex EEG signals, and is divided into several nonoverlapping specific duration windows by the cortex eeg collection system.Cortex EEG signals in each window is asked for reference value, with the intermediate value between its maximum and minima, as the reference value of this cortex EEG Processing.

Step (2) is to the cortex EEG signals in each window, carry out respectively the bipolar processes of amplitude: the cortex EEG signals value in each window is deducted the reference value that step (1) is tried to achieve, acquisition has ambipolar cortex EEG signals, makes it satisfy the neuron models input signal and has ambipolar requirement.

The value of feedback of the cortex EEG signals that step (3) obtains step (2), the noise signal of interpolation and output signal is as the input signal based on FHN neuron Feedback network model, utilize the accidental resonance effect of FHN neuron Feedback network model, obtain the cortex brain electrical output signal that strengthens.

Step (4) under the varying strength that adds noise signal, is calculated respectively the signal to noise ratio of FHN neuron Feedback network model response in the step (3).Utilize random resonance mechanism, when signal to noise ratio reached maximum, the optimum that the cortex EEG signals will obtain under the signal to noise ratio evaluation index this moment strengthened, with this cortex EEG signals as output signal.

The reference value that step (5) is asked for above-mentioned output signal and step (1) is sued for peace, and former amplitude scope is returned in inverse mapping, thereby obtains the cortex EEG signals of the enhancing after the amplitude reduction.

Beneficial effect of the present invention:

1, because the cortex EEG signals has transient characteristic, the cortex EEG signals that the present invention arrives dynamic acquisition, its intermediate value is asked in setting in short-term long window, obtains the reference value that follow-up bipolarity mapping is processed, this reference value has dynamic characteristic, and the real-time processing that utilizes the transition EEG signals is arranged.

2, the present invention is based on the random resonance mechanism of FHN neuron Feedback network model, different from traditional signal Enhancement Method based on noise filtering, its passive power conversion with noise is the positive energy of signal, thereby realizes the enhancing of faint cortex EEG signals.

3, the present invention has given up the abstract models such as bistable system commonly used in the accidental resonance of EEG signals strengthens, but adopts the neuron models that meet true neuron electrophysiological characteristics; Network models in the unidirectional connection of neuron has increased feedback element simultaneously, more meets the interconnected relationship between the neuron in the nervous system, is conducive to improve the stability in the faint cortex EEG signals enhancing process.

Description of drawings

Fig. 1 is FHN neuron feedback double-layer network model structure schematic diagram.

The specific embodiment

Step (1) is used cortex eeg signal acquisition system, gathers one section continuous cortex EEG signals, and it is divided into several nonoverlapping windows, and the window duration is designated as

, wherein NSampling number in the expression window, TThe expression sampling period.Therefore the cortex EEG signals in the window can be designated as

, (

).It is asked for maximum

And minima

, their average as the dynamic benchmark value of this window EEG signals, is designated as

Step (2) is carried out the bipolar processes of amplitude with the cortex EEG signals in each window of step (1).Each cortex EEG signals sampled value that is about in the window deducts respectively the reference value that step (1) is tried to achieve, and obtains to have ambipolar cortex EEG signals

, (

), make it satisfy the input signal bipolarity characteristics of neuron models.

Cortex EEG signals, the white Gaussian noise of interpolation and the value of feedback of output signal that step (3) obtains step (2) are as the input signal of FHN neuron Feedback network model.

Be specifically described as an example of bilayer feedback FHN neural network model example, wherein the model structure schematic diagram as shown in Figure 1, among the figure

Be the cortex EEG signals of current time after the bipolarity mapping is processed;

It is the independent noise item of same noise intensity;

Be the feedback regulation parameter, wherein network system is negative feedback;

The of expression ground floor

The neuronic coefficient of connection of individual neuron and the second layer;

Be of ground floor

Individual FHN neuron;

Be second layer FHN neuron;

Be the neuronic output membrane voltage of the second layer, i.e. output signal.

Mathematical model is suc as formula shown in (1) and the formula (2):

Ground floor:

(1)

The second layer:

(2)

In formula (1) and the formula (2), Be the sampling period;

Be output signal;

Be time constant, determined neuronic ignition rate;

Be marginal value, impel neuron regularly to light a fire;

For the signal level average with Difference;

,

Be the equation group constant;

Be the input signal of bilayer feedback FHN neural network model,

,

Be the cortex EEG signals after the bipolarity mapping processing, The noise item in the input signal, usually by average be 0, auto-correlation function is

White Gaussian noise simulated, wherein

Be noise intensity,

The expression impulse function;

Be the feedback regulation parameter; Be the first floor iIndividual neuronic output membrane voltage;

Be the neuronic output membrane voltage of the second layer, i.e. output signal;

Slowly become the recovery variable for the second layer is neuronic; Coefficient of connection is

Utilize the value of feedback of the noise signal of cortex EEG signals to be strengthened, interpolation and response as excitation, random resonance mechanism by FHN neuron Feedback network model, thereby realize the enhancing of faint cortex EEG signals, obtain the cortex brain electrical output signal that strengthens.

Step (4) is for the noise signal in the step (3)

, get respectively varying strength DNoise figure.Calculate the signal to noise ratio of FHN neuron Feedback network model response under the varying strength noise.According to random resonance mechanism, in the certain noise strength range, along with the increase of noise intensity, the response signal to noise ratio is with monotone increasing; And when noise intensity was increased to certain value, if continue to increase noise intensity, the response signal to noise ratio can descend on the contrary, until noise floods signal fully.Therefore when the response signal to noise ratio reached maximum, the optimum that the cortex EEG signals will obtain on the signal to noise ratio meaning this moment strengthened.Wherein signal to noise ratio is defined as:

(3)

Wherein,

,

Represent respectively the cortex EEG signals output signal of corresponding certain window duration in power spectral density and the noise signal of interpolation, the unit of signal to noise ratio is decibel (dB).Consider the randomness of noise, therefore in power spectral density is calculated, adopt power spectral density cumulative mean method.Namely under the cortex EEG signals and noise intensity effect of identical input, repeat to ask for the response value of model, to every group of response difference rated output spectrum density, again the power spectral density of all response values is carried out cumulative mean.

The reference value that step (5) is asked for above-mentioned output signal and step (1) is sued for peace, and former amplitude scope is returned in inverse mapping, namely obtains the cortex EEG signals of the enhancing of amplitude reduction.

Claims

1. the feedback system accidental resonance Enhancement Method of a cortex EEG signals is characterized in that the method comprises the steps:

Step 1, by the cortex eeg collection system, obtain the cortex EEG signals, and be divided into several nonoverlapping specific duration windows, cortex EEG signals in each window is asked for reference value, the method of asking for is with the maximum of the cortex EEG signals in each window and the intermediate value between minima, as the described reference value of this cortex EEG Processing;

Step 2, to the cortex EEG signals in each window, carry out respectively the bipolar processes of amplitude, specifically: the cortex EEG signals value in each window is deducted the reference value that step 1 is tried to achieve, acquisition has ambipolar cortex EEG signals, makes it satisfy the neuron models input signal and has ambipolar requirement;

The noise signal of step 3, the cortex EEG signals that step 2 is obtained, interpolation and the value of feedback of output signal are as the input signal based on FHN neuron Feedback network model, utilize the accidental resonance effect of FHN neuron Feedback network model, obtain the cortex brain electrical output signal that strengthens;

Step 4, in the step 3, adding under the varying strength of noise signal, calculate respectively the signal to noise ratio of FHN neuron Feedback network model response; Utilize random resonance mechanism, when signal to noise ratio reached maximum, the optimum that the cortex EEG signals will obtain under the signal to noise ratio evaluation index this moment strengthened, with this cortex EEG signals as output signal;

Step 5, the reference value that the output signal in the step 4 and step 1 are asked for are sued for peace, and former amplitude scope is returned in inverse mapping, thereby obtains the cortex EEG signals of the enhancing after the amplitude reduction.